Skip to main content
Celestibia Solutions

Real-Time Fraud Detection with Streaming Analytics for a FinTech

A payments and digital banking platform replaced batch, rule-based fraud checks with Kafka/Kinesis streaming and ML scoring on Amazon EKS: sub-second detection and 99.99% availability at peak.

Discuss a similar projectRead the full write-up
  • Data analytics
  • FinTech
  • AWS
TransactionsKafka / KinesisML scoringon Amazon EKSApproveFlagBlockSub-second decisions99.99% availability at peak
Each transaction is scored and approved, flagged or blocked in under a second.
fraud detection for high-risk transactions
Sub-second
availability during peak transaction volumes
99.99%
false-positive rate for genuine customers
Lower

At a glance

Engagement
Data analytics
Industry
FinTech
Platform
AWS
Region
Not published
Client
Payments and digital banking company, name withheld
Technologies
12 in the delivered stack

Services delivered

Before and after

A platform processing millions of transactions a day relied on slow, reactive rule-based fraud detection. False positives frustrated genuine customers, and batch analytics delivered insight too late.

Before

  • Fraud had to be detected in real time, within milliseconds, not after a batch run.
  • Analytics had to absorb transaction spikes without degrading.
  • Sensitive financial data needed encryption, compliance and auditability throughout.

After

  • Sub-second fraud detection for high-risk transactions
  • 99.99% availability during peak transaction volumes
  • Lower false-positive rate for genuine customers

The solution

Solution path

  1. Transactions
  2. Kafka / Kinesis
  3. ML scoring on EKS
  4. Decision engine
  5. Approve · flag · block

How it was delivered

  1. Phase 1

    Built an event-driven streaming pipeline on Kafka and Amazon Kinesis to ingest transaction events with sub-second latency.

  2. Phase 2

    Ran microservice analytics on Amazon EKS with ML models for behavioural and anomaly detection, feeding real-time scoring and decision engines that approve, flag or block.

  3. Phase 3

    Added auto-scaling, encryption at rest and in transit, monitoring and alerting for fraud events, and DevSecOps controls for compliance.

Technology stack

  • Apache Kafka
  • Amazon Kinesis
  • Amazon EKS
  • Docker
  • Microservices
  • Machine learning models
  • Amazon RDS PostgreSQL
  • Amazon DynamoDB
  • Amazon S3
  • AWS Lambda
  • AWS IAM
  • Amazon CloudWatch

Questions about this project

How does the platform detect fraud in real time?

Transaction events stream through Kafka and Amazon Kinesis to analytics microservices on Amazon EKS, where machine learning models score each transaction and a decision engine approves, flags or blocks it within a second.

How does it cope with transaction spikes?

The analytics services auto-scale on Kubernetes, which kept the platform at 99.99% availability during peak transaction volumes without manual intervention.

Related services

Data Engineering ServicesBatch and streaming pipelines, warehouses, lakehouses, quality and governance.AI & Machine Learning SolutionsModel development, MLOps, computer vision, NLP and AI infrastructure.DevSecOps ServicesSecurity testing, scanning, secrets and policy as code inside CI/CD.

Related technical articles

Case study · InsightsFinTech Data Analytics: Real-Time Fraud Detection on Cloud-Native Architecture (opens on insights.celestibia.com)
Read the full case study on Celestibia Insights

Planning something similar?

Whether you're migrating, cutting costs, securing your infrastructure or exploring startup programs, start with the Cloud Readiness Survey.

Discuss your projectMore case studies

What happens next

  1. Answer six questions about your cloud setup, spend and priorities. It takes about three minutes.
  2. A Celestibia engineer reviews your answers within one business day.
  3. You get tailored recommendations and the AWS programs you are likely to qualify for.

Monday to Friday, 10 AM to 7 PM IST